3c057cb2b41f22c0e740974d7a428918-Paper.pdf

One important limitation of VAEs is the strong prior assumption that latent representations learned by the model followasimpleuni-modal Gaussian distribution. Further,thevariational training procedure poses considerable practical challenges. Recently proposed regularized autoencoders offeradeterministic autoencoding framework, that simplifies the original VAE objective and is significantly easier to train. Since these models only provide weak control over the learned latent distribution, they require an ex-post density estimation step to generate samples comparable to those of VAEs. Inthispaper,wepropose asimple andend-to-end trainable deterministic autoencoding framework, that efficiently shapes the latent space of the model during training and utilizes the capacity of expressive multi-modal latent distributions. The proposed training procedure provides direct evidence if the latent distribution adequately captures complexaspects oftheencoded data.